In order to analyse the results from an on-farm survey in Burkina Faso, a model for grain yield analysis was calibrated using data from the survey. Yield was broken down into four components, each representing one phase of the growth cycle. In the model, the maximum value for each component (M) is defined by the cultivar in the region of study, and from the values that the previous components have reached. These values establish the competition level for resources that a growing component undergoes. Competition begins beyond a critical value (L) of plant density (NP). An upper boundary line is defined on the base of NP for each component. A maximum grain yield (YM/NP) is evaluated for each NP value. The greatest maximum yield (Y GMAX) is considered as a varietal characteristic under the given conditions. The model was applied to two varietal groups (local and selected varieties). It indicated that selected varieties had higher Y GMAX' mainly due to different competition critical limit values (L) and not to varietal maxima (M). For each group of varieties, the later in the cycle a component is determined, the lower the competition limit YM/NP increases with NP up to a first competition limit. Subsequently, adjustments occur between successive components, thereby stabilizing Y M/NP (= Y GMAX) across levels of NP. Finally, when the fraction of fertile plants is limited by NP, Ym/Np decreases against NP. The lowest NP enabling Y GMAX'while providing the greatest potential for adjusting yield components, gives the best yields. This hypothesis was verified under field conditions. The range of population densities associated with Y GMAX and greatest potential for adjustment were also the most frequently observed on farm, indicating that farmers were probably familiar with the effects of population density on yield potential.